Workforce management without prediction models is like a phone without apps. That is the opinion of Jan van den Berg, manager BI & Control GV at Quion. ‘Only when you have good software can you really do fun things with it.’ With Pipple as a partner, Quion has started a process to derive more predictive power from their figures.
If anyone is good with numbers, it’s Quion. They take the processes involved in taking out and managing mortgages for lenders. And they advise them on how to streamline accounting. But internally, the management of data could use a boost.
A statistical candy store revealed itself when they came up with cases with Pipple where prediction models could be worth their weight in gold. All kinds of great initiatives were put on the table, from a long-term absenteeism warning system and a fraud detection model to automatic credit risk assessment. It was decided to start with a model to predict the workload for back office, mid-office and telephony on a weekly basis. That would have helped the operation the most.
Jan had already had a student experiment with a model for this. ‘As a result, we knew the relevant parameters and knew that the quality of the required data was in order’, Jan explains. ‘A further developed model would, in addition to being useful, also be a nice buy-in for other models on our wish list.’
Quion brought pipple consultant Vera van der Lelij in-house for a longer period of time. Jan: ‘Her start with us was as I had hoped from Pipple: she didn’t need lengthy conversations, we didn’t have to ‘live through’ the problem together first and she didn’t do endless sending about what we had to do. She just went to work.’ The process Vera is responsible for is simple: she spars with a mathematically savvy team member of Quion who knows a lot about the data and then models. A third team member deploys the new model to the data warehouse. Is the result insufficient? Then the whole thing starts all over again. Vera was soon able to show that the original model that Quion assumed deviated considerably from reality. Not something that Jan and his people let themselves be knocked out of. ‘Drawing up a model can only be done iteratively. Failure is part of the process.’
No button course
Pipple is driven to get the numbers right, Jan notices. ‘They only really like a model if it is 99% reliable. That’s their mathematical nature.’ Jan lowers the bar. “Predictions are just predictions. Sometimes they don’t come out. If we tolerate a deviation of 10% in this specific model, we plan at most 2 FTE too many or too few. That is already a huge win for us.’
What Jan does find very important is that Pipple lets his team understand the model. He is not satisfied with a button course to roll out numbers. ‘I want us to be able to adjust the model ourselves if necessary if we maintain the models independently.’ Fortunately, Vera constantly ensures that the necessary quarters fall, he notes with satisfaction.
As far as Jan is concerned, Pipple will remain at home with Quion for the time being. Because just like Quion, they are go-getters and the value grows as Pipple gets to know the mortgage world better. Moreover: ‘One app on your phone is of course not nearly enough.’